Chapter 10: I/O Subsystems (2)

Transcription

1 ADRIAN PERRIG & TORSTEN HOEFLER ( ) Networks and Operating Systems Chapter 10: I/O Subsystems (2) BE CAREFUL WITH I/O DEVICES!

2 Our Small Quiz True or false (raise hand) Open files are part of the process address-space Unified buffer caches improve the access times A partition table can unify the view of multiple disks Unix enables to bind arbitrary file systems to arbitrary locations The virtual file system interface improves modularity of OS code Programmed I/O is efficient for the CPUs DMA enables devices to access virtual memory of processes IOMMUs enable memory protection for devices IOMMUs improve memory access performance First level interrupt handlers process the whole request from the hardware Software interrupts reduce the request processing latency Deferred procedure calls execute second-level interrupt handlers 2

3 The I/O subsystem spcl.inf.ethz.ch

4 Generic I/O functionality Device drivers essentially move data to and from I/O devices Abstract hardware Manage asynchrony OS I/O subsystem includes generic functions for dealing with this data Such as

5 The I/O Subsystem Caching - fast memory holding copy of data Always just a copy Key to performance Spooling - hold output for a device If device can serve only one request at a time E.g., printing

6 The I/O Subsystem Scheduling Some I/O request ordering via per-device queue Some OSs try fairness Buffering - store data in memory while transferring between devices or memory To cope with device speed mismatch To cope with device transfer size mismatch To maintain copy semantics

7 Naming and Discovery What are the devices the OS needs to manage? Discovery (bus enumeration) Hotplug / unplug events Resource allocation (e.g., PCI BAR programming) How to match driver code to devices? Driver instance driver module One driver typically manages many models of device How to name devices inside the kernel? How to name devices outside the kernel?

8 Matching drivers to devices Devices have unique (model) identifiers E.g., PCI vendor/device identifiers Drivers recognize particular identifiers Typically a list Kernel offers a device to each driver in turn Driver can claim a device it can handle Creates driver instance for it.

9 Naming devices in the Unix kernel (Actually, naming device driver instances) Kernel creates identifiers for Block devices Character devices [ Network devices see later ] Major device number: Class of device (e.g., disk, CD-ROM, keyboard) Minor device number: Specific device within a class

10 Unix Block Devices Used for structured I/O Deal in large blocks of data at a time Often look like files (seekable, mappable) Often use Unix shared buffer cache Mountable: File systems implemented above block devices

11 Character Devices Used for unstructured I/O Byte-stream interface no block boundaries Single character or short strings get/put Buffering implemented by libraries Examples: Keyboards, serial lines, mice Distinction with block devices somewhat arbitrary

12 Naming devices outside the kernel Device files: special type of file Inode encodes <type, major num, minor num> Created with mknod() system call Devices are traditionally put in /dev /dev/sda First SCSI/SATA/SAS disk /dev/sda5 Fifth partition on the above /dev/cdrom0 First DVD-ROM drive /dev/ttys1 Second UART

13 Pseudo-devices in Unix Devices with no hardware! Still have major/minor device numbers. Examples: /dev/stdin /dev/kmem /dev/random /dev/null /dev/loop0 etc.

14 Old-style Unix device configuration All drivers compiled into the kernel Each driver probes for any supported devices System administrator populates /dev Manually types mknod when a new device is purchased! Pseudo devices similarly hard-wired in kernel

15 Linux device configuration today Physical hardware configuration readable from /sys Special fake file system: sysfs Plug events delivered by a special socket Drivers dynamically loaded as kernel modules Initial list given at boot time User-space daemon can load more if required /dev populated dynamically by udev User-space daemon which polls /sys

16 Interface to network I/O spcl.inf.ethz.ch

17 Unix interface to network I/O You already know the data path BSD sockets bind(), listen(), accept(), connect(), send(), recv(), etc. Have not yet seen: Device driver interface Configuration Routing

18 Software routing User space Kernel space Routing protocol messages Routing daemon Protocol stack Routing control FIB (forwarding information base) OS protocol stacks include routing functionality Routing protocols typically in a user-space daemon Non-critical Easier to change Forwarding information typically in kernel Needs to be fast Integrated into protocol stack Network

19 Network stack implementation spcl.inf.ethz.ch

20 Networking stack Probably most important peripheral GPU is increasingly not a peripheral Disk interfaces look increasingly like a network But NO standard OS textbook talks about the network stack! Good references: The 4.4BSD book (for Unix at least) George Varghese: Network Algorithmics (up to a point)

21 Receiving a packet in BSD Application Application Stream socket Datagram socket TCP UDP ICMP IP Receive queue Kernel Network interface

22 Receiving a packet in BSD Application Application Stream socket Datagram socket TCP UDP ICMP IP Kernel Network interface Receive queue 1. Interrupt 1.1 Allocate buffer 1.2 Enqueue packet 1.3 Post s/w interrupt

23 Receiving a packet in BSD Application Application Stream socket Datagram socket TCP UDP ICMP IP Receive queue 2. S/W Interrupt High priority Any process context Defragmentation TCP processing Enqueue on socket Kernel Network interface

24 Receiving a packet in BSD Application Application Stream socket Datagram socket 3. Application Copy buffer to user space Application process context TCP UDP ICMP IP Receive queue Kernel Network interface

25 Sending a packet in BSD Application Application Stream socket Datagram socket TCP UDP ICMP IP Receive queue Kernel Network interface

26 Sending a packet in BSD Application Application Stream socket 1. Application Datagram Copy from user space to buffer socket Call TCP code and process Possible enqueue on socket queue TCP UDP ICMP IP Receive queue Kernel Network interface

27 Sending a packet in BSD Application Application Stream socket Datagram socket TCP UDP ICMP IP Receive queue 2. S/W Interrupt Any process context Remaining TCP processing IP processing Enqueue on i/f queue Kernel Network interface

28 Sending a packet in BSD Application Application Stream socket Datagram socket TCP UDP ICMP IP Kernel Network interface Receive queue 3. Interrupt Send packet Free buffer

29 The TCP state machine Close / FIN SYN_rcvd Closed Passive Open Close Listen SYN / SYNACK SYN / SYNACK ACK Established Close Send / SYN SYNACK / ACK Active open / SYN SYN_sent Close / FIN FIN / ACK FIN_wait_1 FIN / ACK Close_wait ACK Close / FIN FIN_wait_2 Closing Last_Ack FIN / ACK ACK Time_wait Timeout after 2 segment lifetimes ACK Closed

30 OS TCP state machine More complex! Also needs to handle: Congestion control state (window, slow start, etc.) Flow control window Retransmission timeouts Etc. State transitions triggered when: User request: send, recv, connect, close Packet arrives Timer expires Actions include: Set or cancel a timer Enqueue a packet on the transmit queue Enqueue a packet on the socket receive queue Create or destroy a TCP control block

31 In-kernel protocol graph TCP IP UDP ICMP Interfaces can be standard (e.g. X- kernel, Windows) or protocol-specific (e.g. Unix) e.g. Tunneling ARP Ethernet Ethernet device

32 Protocol graphs Graph nodes can be: Per-protocol (handle all flows) Packets are tagged with demux tags Per-connection (instantiated dynamically) Multiple interfaces as well as connections Ethernet Ethernet bridging IP IP IP routing

33 Memory management spcl.inf.ethz.ch

34 Memory management Problem: how to ship packet data around Need a data structure that can: Easily add, remove headers Avoid copying lots of payload Uniformly refer to half-defined packets Fragment large datasets into smaller units Solution: Data is held in a linked list of buffer structures

35 BSD Unix mbufs (Linux equivalent: sk_buffs) next offset length type Data (112 bytes) next object

36 BSD Unix mbufs (Linux equivalent: sk_buffs) next offset length type Next mbuf for this object Type: DATA 36 bytes Data (112 bytes) 24 bytes next object Next object in a list

37 Case Study: Linux 3.x Implementing a simple protocol over Ethernet Why? You want to play with networking equipment (well, RAW sockets are easier) You want to develop specialized protocols E.g., application-specific TCP E.g., for low-latency cluster computing You ll understand how it works!

38 Sending Data in Linux 3.x Socket char* tcp_send_msg struct sk_buff tcp_transmit_skb struct sk_buff, TCP Header ip_queue_xmit ip_fragment ip_route_output_flow dev_queue_xmit Many layers Most use the sk_buf struct ip_forward Driver

39 Register a receive hook Fill packet_type struct:.type = your ethertype.func = your receive function Receive handler recv_hook(...) Gets sk_buff, packet_type, net_device,... Called for each incoming frame Reads skb->data field and processes protocols Receive hook table: 0x0800 IPv4 hdlr. 0x8864 PPPOE hdlr. 0x8915 RoCE hdlr.

40 Interaction with applications Socket Interface Need to implement handlers for connect(), bind(), listen(), etc. Call sock_register(struct net_proto_family*) Register a protocol family Enables user to create socket of this type

41 Anatomy of struct sk_buff Called skb in Linux jargon Allocate via alloc_skb() (or dev_alloc_skb() if in driver) Free with kfree_skb() (dev_kfree_skb()) Use pskb_may_pull(skb, len) to check if data is available skb_pull(skb, len) to advance the data pointer... it even has a webpage!

42 SKB Fields Double-linked list, each skb has.next/.prev.data contains payload (size of data field is set by skb_alloc).sk is the socket this skb is owned by.mac_header,.network_header,.transport_header contain headers of various layers.dev is the device this skb uses member fields total

43 Case Study: TCP Fragmenting Linux <2.0.32: Two fragments: #1 Offset: 0 Length: 100 #2 Offset 100 Length: 100 // Determine the position of this fragment. end = offset + iph->tot_len - ihl; #1: 100, #2: 200 // Check for overlap with preceding fragment, and, if needed, // align things so that any overlaps are eliminated. if (prev!= NULL && offset < prev->end) { i = prev->end - offset; offset += i; /* ptr into datagram */ ptr += i; /* ptr into fragment data */ } // initialize segment structure fp->offset = offset; #1: 0, #2: 100 fp->end = end; #1: 100, #2: 200 fp->len = end - offset; #1: 100, #2: // collect multiple such fragments in queue // process each fragment if(count+fp->len > skb->len) { error_to_big; } memcpy((ptr + fp->offset), fp->ptr, fp->len); count += fp->len; fp = fp->next;

44 Case Study: TCP Fragmenting Linux <2.0.32: Two fragments: #1 Offset: 0 Length: 100 #2 Offset 10 Length: 20 // Determine the position of this fragment. end = offset + iph->tot_len - ihl; #1: 100, #2: 30 // Check for overlap with preceding fragment, and, if needed, // align things so that any overlaps are eliminated. if (prev!= NULL && offset < prev->end) { i = prev->end - offset; #2: =90 offset += i; /* ptr into datagram */ #2: 100 ptr += i; /* ptr into fragment data */ } // initialize segment structure fp->offset = offset; fp->end = end; fp->len = end - offset; #1: 0, #2: 100 #1: 100, #2: 30 #1: 100, #2: // collect multiple such fragments in queue // process each fragment if(count+fp->len > skb->len) { error_to_big; } memcpy((ptr + fp->offset), fp->ptr, fp->len); count += fp->len; fp = fp->next; (size_t)-70 =

45 Case Study: TCP Fragmenting Linux <2.0.32: Two fragments: #1 Offset: 0 Length: 100 #2 Offset 10 Length: 20 // Determine the position of this fragment. end = offset + iph->tot_len - ihl; #1: 100, #2: 30 // Check for overlap with preceding fragment, and, if needed, // align things so that any overlaps are eliminated. if (prev!= NULL && offset < prev->end) { i = prev->end - offset; #2: =90 offset += i; /* ptr into datagram */ #2: 100 ptr += i; /* ptr into fragment data */ } // initialize segment structure #1: 0, #2: 100 fp->offset = offset; fp->end = end; #1: 100, #2: 30 fp->len = end - offset; #1: 100, #2: // collect multiple such fragments in queue // process each fragment if(count+fp->len > skb->len) { error_to_big; } memcpy((ptr + fp->offset), fp->ptr, fp->len); count += fp->len; fp = fp->next; (size_t)-70 =

46 that s so last century! spcl.inf.ethz.ch

47 Performance issues spcl.inf.ethz.ch

48 Life Cycle of an I/O Request Request I/O User process I/O complete Can satisfy request? System call No Yes Send request to driver Block process if needed Issue commands to device Block until interrupted I/O subsystem Unfortunately, this is a bit of a convenient fiction Device driver Transfer data to/from user space, Return completion code Demultiplex I/O complete Determine source of request Return from system call Issue interrupt when I/O completed Interrupt handler Physical device Handle interrupt Signal device driver Interrupt I/O complete Generate Interrupt Time

49 Consider 10 Gb/s Ethernet spcl.inf.ethz.ch

50 At full line rate for 1 x 10Gb port ~1GB (gigabyte) per second ~ 700k full-size Ethernet frames per second At 2GHz, must process a packet in 3000 cycles This includes: IP and TCP checksums TCP window calculations and flow control Copying packet to user space

51 A few numbers L3 cache miss (64-byte lines) 300 cycles At most 10 cache misses per packet Note: DMA ensures cache is cold for the packet! Interrupt latency 500 cycles Kernel entry/exit Hardware access Context switch / DPC Etc.

52 Plus You also have to send packets. Card is full duplex can send at 10Gb/s You have to do something useful with the packets! Can an application can make use of 1.5kB of data every 1000 machine cycles or so? This card has two 10Gb/s ports.

53 And Plus And if you thought that was fast Mellanox 100 Gb/s Adapter Impossible to use without advanced features RDMA SR-IOV TOE Interrupt coalescing

54 What to do? TCP offload (TOE) Put TCP processing into hardware on the card Buffering Transfer lots of packets in a single transaction Interrupt coalescing / throttling Don t interrupt on every packet Don t interrupt at all if load is very high Receive-side scaling Parallelize: direct interrupts and data to different cores

55 Linux New API (NAPI) Mitigate interrupt pressure 1. Each packet interrupts the CPU 2. Vs. CPU polls driver NAPI switches between the two! NAPI-compliant drivers Offer a poll() function Which calls back into the receive path

56 Linux NAPI Balancing Driver enables polling with netif_rx_schedule(struct net_device *dev) Disables interrupts Driver deactivates polling with netif_rx_complete(struct net_device *dev) Re-enable interrupts. but where does the data go???

57 Buffering Key ideas: Decouple sending and receiving Neither side should wait for the other Only use interrupts to unblock host Batch together requests Spread cost of transfer over several packets

58 Producer-consumer buffer descriptor rings Free descriptors Consumer pointer Producer pointer Full descriptors Descriptor format Physical address Size in bytes Misc. flags

59 Buffering for network cards Producer, consumer pointers are NIC registers Transmit path: Host updates producer pointer, adds packets to ring Device updates consumer pointer Receive path: Host updates consumer pointer, adds empty buffers to ring Device updates producer pointer, fills buffers with received packets. More complex protocols are possible

60 Example transmit state machine Sends packet; More packets in descriptor ring Sends last packet; None left in descriptor ring Running Idle Sends packet; Ring occupancy below threshold Host updates producer pointer; Ring now full Host updates producer pointer Running; host blocked Sends packet; Ring still nearly full

61 Transmit interrupts Ring empty all packets sent device going idle Ring occupancy drops host can now send again device continues running Sends packet; More packets in descriptor ring Sends packet; Ring occupancy below threshold Running Sends last packet; None left in descriptor ring Host updates producer pointer; Ring now full Idle Host updates producer pointer Exercise: devise a similar state machine for receive! Running; host blocked Sends packet; Ring still nearly full

62 Buffering summary DMA used twice Data transfer Reading and writing descriptors Similar schemes used for any fast DMA device SATA/SAS interfaces (such as AHCI) USB2/USB3 controllers etc. Descriptors send ownership of memory regions Flexible many variations possible: Host can send lots of regions in advance Device might allocate out of regions, send back subsets Buffers might be used out-of-order Particularly powerful with multiple send and receive queues

63 Receive-side scaling Insight: Too much traffic for one core to handle Cores aren t getting any faster Must parallelize across cores Key idea: handle different flows on different cores But: how to determine flow for each packet? Can t do this on a core: same problem! Solution: demultiplex on the NIC DMA packets to per-flow buffers / queues Send interrupt only to core handling flow

64 Receive-side scaling Received packet Hash of packet header IP src + dest TCP src + dest Etc. Flow table pointer Flow state: Ring buffer Message-signalled interrupt DMA address Core to interrupt

65 Receive-side scaling Can balance flows across cores Note: doesn t help with one big flow! Assumes: n cores processing m flows is faster than one core Hence: Network stack and protocol graph must scale on a multiprocessor. Multiprocessor scaling: topic for later

66 Tomorrow Virtual machines Multiprocessor operating systems